Diceware passphrase lengths are on the rise - up to six or seven words now. The old adage that passphrases are easier to remember may be true for shorter phrases, but six truly random words can be tough to remember. On the other hand, full sentences may be easier for some to remember.
Take for example the Diceware-generated passphrase
tracy optic renown acetic sonic kudo. We could turn that into a (nonsensical) sentence such as
Tracy's optics were renowned, but her acetic sonic cost her kudos.
The Diceware passphrase has an entropy of 77.4 if the attacker knows you're using six Diceware words (12.9 per word), and 107.219 (according to this calculator) if they don't. The sentence form has an entropy (according to the calculator) of 255.546. However, it's not fully random any more, which is supposed to be one of the big benefits of the Diceware approach.
Assuming that the attacker somehow knows that you're using this method of passphrase generation, does the sentence form decrease the security of the passphrase in any way? For example, perhaps they can use some kind of analysis of English sentence structure to narrow down their required guesses?
Assuming the answer to the above is "No, sentence form does not decrease security," then here's another consideration:
One benefit of the sentence format is that it's very long and includes non-alphabetical characters (eg. the apostrophe and comma). However, that's a definite downside when trying to type it on a mobile device. Say we shorten the Diceware phrase to three words -
tracy optic renown - and then turn that into an
[a-z] sentence -
tracy is optically renowned or perhaps
tracy is optically renowned worldwide (to further distinguish it from the Diceware wordlist).
If we were to use three Diceware words and the attacker knows we're using Diceware then we have an entropy of 38.7. However,
tracy is optically renowned worldwide is 100.504 bits of entropy according to the calculator.
Given the differences between the three word Diceware phrase and the short sentence form, which entropy calculation is more accurate - the Diceware calculation (ie. the differences are too slight to matter) or the calculator's calculation (dictionary/brute-force/etc.)?
Note: assume that any length or combination of characters is acceptable for the password